A method for explaining model predictions by measuring how much each input feature contributes to the final output.